The chain
$0.184 for the whole chain, at each model’s list price:
- UI sounds x 6: $0.006 ($0.002 per second, 6 jobs)
- Footsteps x 12: $0.012 ($0.002 per second, 12 jobs)
- Hits and impacts x 8: $0.016 ($0.002 per second, 8 jobs)
- Swings and whooshes x 4: $0.0824 ($0.0206 per sound, 4 jobs)
- Pickups x 4: $0.008 ($0.002 per second, 4 jobs)
- Ambience loop: $0.06 ($0.002 per second)
summer api estimate <model> --input '<json>', or POST /v1/estimate with the same body as the job. Prices come from GET /v1/models and change only when the provider’s price does.
That is 35 files. Sound is the cheapest part of a game to make well, so make three takes of anything the player hears often.
Before you start
Sign in once, so your agent or your terminal can run jobs on your Summer account:- The palette. Two or three material words for your world: “warm wood and soft cloth” for a cozy farming game, “glass and thin metal” for a sci-fi menu. Every UI and pickup prompt uses them.
- The tail rule. Every one-shot is dry:
close-mic, dry, no reverb. The room comes from the engine, not the file. - The closing words.
no music, no voicesat the end of every prompt, word for word. Without them a model can add a melody under a pickup or a grunt under a hit.
Step 1: UI sounds
UI sounds play hundreds of times an hour, right next to the player’s finger. They must start the instant the button goes down and stop before the next press. Ask for 0.5 seconds, the shortest length, and describe one physical event: “a soft wooden toggle pressed once” beats “a nice menu click”.
Hover is the quietest, confirm the brightest. If the error sound is louder than the click, the player feels punished; lower it in the mix, not by rewriting the prompt.
Step 2: footsteps on four materials
Footsteps are the sound players hear most. One file per surface on repeat sounds mechanical within ten steps. So make three takes of each surface by running the same prompt three times: each run is a different performance. Name the shoe and the surface, and ask for one step only.on dry stone tiles, on wet grass, on old wooden floorboards, on loose gravel. Keep the boot, the rhythm words and the closing words the same, so the character sounds like one person walking across different floors. A step is about 0.5 seconds; ask for more and the model fills the rest with a second step or a scuff you then have to trim.
Step 3: hits and impacts
A hit has two parts: the transient, the sharp first moment that tells the player “contact”, and the tail, which tells them what was hit. Describe both, and give the tail a length: “short ringing tail that dies in half a second”. One second is enough for almost every impact.a blade cuts into a thick leather sack, wet slice), punch (a bare fist hits a heavy leather bag, deep thud with a slap on top, no tail), arrow into wood, a body falling on dirt, a crate breaking, a shield block, and a heavy stomp. Name materials, never feelings: epic, powerful and cool make the result random.
Step 4: swings and whooshes
Movement sounds fail in one common way: you ask for a sword swing and get a clang. Sayair movement only, no metal ring, no impact, and play the impact as its own sound at the moment of contact. That split also lets a miss sound like a miss.
Step 5: pickups
Pickups are rewards, so they should be short, bright and slightly musical: a clink, a pop, a small upward sparkle. Keep them in the same palette as the UI, because players hear both in the same second.a glass bottle uncorked with a short pop and a soft liquid swirl), key (a small iron key lifted from stone, light jingle) and power-up (a rising bright shimmer, half a second, ends on a clean high note). For a level-up or victory fanfare, use a music job instead: see music for your game.
Step 6: the ambience loop
Ambience is the air of a place, and it is a different job from a sound effect. It plays for minutes under everything, so it must be boring: sparse, distant, low, with no event that pulls attention. Ask for the longest length, 30 seconds, and setloop: true. Sound Effects v2 is the only sound model that makes seamless loops.
no loud events at the start or end, because a bird call right at 0:00 or 0:29 makes the loop point easy to hear. For music-like pads under a menu, use a music job; the sound model renders texture, the music model renders melody. The ambient bed skill has prompts for twelve places, from dungeons to spaceships.
Run the pack as one batch
Do not send 35 single runs. Put every job in one JSONL file, one line per sound, and send it as a batch: it is priced up front, runs all or none, and comes back in the order you wrote it.sfx-names.txt, with one name per line in the same order (ui_click, step_stone_01, step_stone_02, swing_sword_01, amb_forest_dawn_loop). Then run the batch and rename the files:
summer_api_batch({ jobs: [...], out: "audio/sfx" }). A batch takes up to 500 jobs, so a whole game’s sound list fits in one (batches).
Naming that scales: <group>_<thing>_<variant>, all lowercase: ui_click, step_grass_02, hit_sword_shield_01, pickup_coin, amb_forest_dawn_loop. Your code can then find every take of a sound by its prefix.
Put it in your game
- Takes and pitch. Never play one footstep file on repeat. In Summer Engine, put the three takes of a surface in an
AudioStreamRandomizer, with random pitch around 0.95 to 1.05 and no repeats. Any engine has the same idea: pick a random take and nudge the pitch a few percent. - Mix by bus, not by file. A starting mix: SFX at 0 dB, UI at -3 dB, ambience at -12 dB, music at -8 dB under voice. Raise a quiet file with its player’s volume, up to about +9 dB, before you regenerate it.
- Room sound. All one-shots are dry, so add reverb per room on the bus: a small room for a hut, a long tail for a cave. The same footstep works in both.
- Trim the front. UI and hit files should start on the sound, not on silence. Trim any lead-in so the click lands with the press.
- Loops. One-shots keep loop off. For the ambience WAV in Summer Engine, set the import loop mode to Forward. If you still hear the seam, cross-fade two players over 1.5 seconds; the ambient bed skill has the script.
- Inside Summer Engine. The sound effect skill makes three versions with three models, lets you pick by ear, and wires the winner to the right node and bus.
Common problems
Questions
What format do the files come in? WAV, from every sound model. WAV loops cleanly and imports into any engine. Can I loop sounds from any sound model? Seamless loops come from Sound Effects v2 withloop: true. The other sound models make one-shots; for a short hum or engine loop, use Sound Effects v2 too.
Which model is best? It changes from one sound to the next. Sound Effects v2 and Sonilo SFX 1.1 are strong on most sounds, Stable Audio 3 SFX on fast movement, and Mirelo SFX 2.0 ($0.01 per second) is worth a test on hero sounds such as a boss roar. For a sound the player hears every second, run the same prompt on two models and pick by ear.
Can it make grunts and screams? Yes, as sounds: male grunt of pain, mid-thirties, sharp, short. For words, use a voice model: see voice lines for game characters.
Can I use the sounds in a commercial game? Yes. Every model on the Summer API allows commercial use of what it makes.